In [14]:
# 查看当前挂载的数据集目录
!ls /home/aistudio/
In [47]:
# 查看个人持久化工作区文件
!ls /home/aistudio/work/
基于GEI的HOG特征，使用SVM分类器分类步态识别¶
采用中科院自动化所GaitDatasetA-silh小型步态识别数据集进行实验分析
将数据集每人序列帧图像平均分为两部分，一部分为训练集，一部分为测试集
对行人二值化序列帧图像求平均得到GEI步态能量图，再采用OpenCV内置HOGDescriptor提取出GEI的HOG特征值，利用HOG特征训练SVM分类器，最终采用模型进行预测。识别率达到99.1667%
图片预处理¶
对人体区域进行裁剪，将人体对称轴放于图片中央。
In [2]:
import cv2 as cv
from matplotlib import pyplot as plt
import numpy as np
import os
%matplotlib inline
def cut_img(img, T_H, T_W):
# 获得最高点和最低点
y = img.sum(axis=1)
y_top = (y != 0).argmax(axis=0)
y_btm = (y != 0).cumsum(axis=0).argmax(axis=0)
img = img[y_top:y_btm + 1, :]
# 如果高比宽要大，用高计算 resize 的比例
_r = img.shape[1] / img.shape[0]
_t_w = int(T_H * _r)
img = cv.resize(img, (_t_w, T_H), interpolation=cv.INTER_CUBIC)
# 获得人的对称轴
sum_point = img.sum()
sum_column = img.sum(axis=0).cumsum()
x_center = -1
for i in range(sum_column.size):
if sum_column[i] > sum_point / 2:
x_center = i
break
h_T_W = int(T_W / 2)
left = x_center - h_T_W
right = x_center + h_T_W
if left <= 0 or right >= img.shape[1]:
left += h_T_W
right += h_T_W
_ = np.zeros((img.shape[0], h_T_W))
img = np.concatenate([_, img, _], axis=1)
img = img[:, left:right]
return img
GEI步态能量图¶
In [3]:
def get_GEI(imgs):
GEI = (imgs[0]/255).astype("uint8")
for img in imgs[1:]:
GEI += (img/255).astype("uint8")
GEI = GEI/len(imgs)
return (GEI*255).astype("uint8")
# GEI = imgs[0].astype("uint8")
# for img in imgs[1:]:
# GEI += img.astype("uint8")
# GEI = GEI/len(imgs)
# return (GEI).astype("uint8")
In [4]:
SZ=20
bin_n = 16 # Number of bins
affine_flags = cv.WARP_INVERSE_MAP|cv.INTER_LINEAR
def get_Hog(img):
# gx = cv.Sobel(img, cv.CV_32F, 1, 0)
# gy = cv.Sobel(img, cv.CV_32F, 0, 1)
# mag, ang = cv.cartToPolar(gx, gy)
# mags.append(mag)
# angs.append(ang)
# bins = np.int32(bin_n*ang/(2*np.pi)) # quantizing binvalues in (0...16)
# bin_cells = bins[:10,:10], bins[10:,:10], bins[:10,10:], bins[10:,10:]
# mag_cells = mag[:10,:10], mag[10:,:10], mag[:10,10:], mag[10:,10:]
# hists = [np.bincount(b.ravel(), m.ravel(), bin_n) for b, m in zip(bin_cells, mag_cells)]
# hist = np.hstack(hists) # hist is a 64 bit vector
# return hist
winSize = (img.shape[1], img.shape[0]) # winSize = (64,64)
blockSize = (8,8) # blockSize = (16,16)
blockStride = (8,8)
cellSize = (8,8)
nbins = 9
derivAperture = 1
winSigma = 4.
histogramNormType = 0
L2HysThreshold = 2.0000000000000001e-01
gammaCorrection = 0
nlevels = 64
hog = cv.HOGDescriptor(winSize,blockSize,blockStride,
cellSize,nbins,derivAperture,
winSigma,histogramNormType,L2HysThreshold,
gammaCorrection,nlevels)
#compute(img[, winStride[, padding[, locations]]]) -> descriptors
winStride = (8,8)
padding = (8,8)
locations = [] # (10, 10)# ((10,20),)
hist = hog.compute(img,winStride,padding,locations)
# hist.dtype = np.float64
return hist
读取GaitDatasetA-silh数据集文件¶
In [5]:
roots = "/home/aistudio/GaitDatasetA-silh/"
cells = []
labels = []
label = os.listdir(roots)
print(label)
# print(label)
for dir_ in label:
dirs = os.listdir(roots + dir_)
for dir__ in dirs:
# print(dir_ + "_" + dir__)
for root_,dirs_,files in os.walk(roots + dir_ + "/" + dir__):
# print(root_)
imgs = []
for file in files:
srcImg = cv.imread(root_ + "/" + file,0)
imgs.append(cut_img(srcImg,128,128))
cells.append(imgs)
labels.append(dir_ + "_" + dir__)
print(labels)
将数据分为训练集与测试集两部分¶
In [6]:
train_cells = []
test_cells = []
for cell in cells:
train_cells.append(cell[0:-1:2])
test_cells.append(cell[1:-1:2])
# train_cells.append(cell[:len(cell)//2])
# test_cells.append(cell[len(cell)//2:])
对训练集构建基于GEI的HOG特征¶
将训练集数据读入，首先提取GEI步态能量图
然后对GEI提取HOG特征，每张GEI的HOG特征为20736 * 1大小
标记240个训练数据，按照人进行编号，每个人12组数据，共20人
输出折线图横轴为HOG数据编号，共240个；纵轴为对应人员编号，共20个
In [7]:
traingeis = []
for train_cell in train_cells:
traingeis.append(get_GEI(train_cell))
trainhogs = []
for traingei in traingeis:
trainhog = get_Hog(traingei)
trainhogs.append(trainhog)
trainData = np.float32(trainhogs).reshape(-1,20736,1)
print(trainData.shape)
responses = np.repeat(np.arange(20),12)[:,np.newaxis]
print(responses.shape)
print(responses.T)
plt.plot(responses)
plt.show()
对测试集构建基于GEI的HOG特征¶
与训练集处理方式相同，每个HOG特征为20736 * 1大小
In [8]:
testgeis = []
for test_cell in test_cells:
testgeis.append(get_GEI(test_cell))
testhogs = []
for testgei in testgeis:
testhog = get_Hog(testgei)
testhogs.append(testhog)
testData = np.float32(testhogs).reshape(-1,20736,1)
print(testData.shape)
采用线性内核函数训练OpenCV内置SVM分类器¶
In [21]:
svm = cv.ml.SVM_create()
svm.setKernel(cv.ml.SVM_LINEAR)
svm.setType(cv.ml.SVM_C_SVC)
svm.setC(5.35)
svm.setGamma(3.58)
svm.train(trainData,cv.ml.ROW_SAMPLE,responses)
svm.save('svm_data.dat')
In [22]:
result = svm.predict(testData)[1].astype("int")
print(result.T)
plt.plot(result)
plt.show()
mask = result==responses
correct = np.count_nonzero(mask)
print(correct*100.0/result.size)
输出测试集与训练集对用HOG,GEI直观对比¶
In [32]:
for i in range(len(trainhogs)):
plt.subplot(141), plt.title("训练集HOG特征"),plt.hist(trainhogs[i])
plt.subplot(142), plt.title("训练集GEI图"),plt.imshow(traingeis[i],cmap='gray')
plt.subplot(143), plt.title("测试集HOG特征"),plt.hist(testhogs[i])
plt.subplot(144), plt.title("测试集GEI图"),plt.imshow(testgeis[i],cmap='gray')
plt.show()
